Claims Browser

Tencent Hunyuan released Hy4-preview, a 770B/49B active, 1M-context open model that immediately looked competitive on coding and SWE-style evals.

"Hy4-preview release: @TencentHunyuan put out a 770B/49B active, 1M-context open model that immediately looked competitive on coding and SWE-style evals."

LeVJEPA claims parity or better than V-JEPA 2 while using 5.6 to 20.8 times less pretraining compute.

"claiming parity or better than V-JEPA 2 at 5.6×–20.8× less pretraining compute"

Automated alignment improvement only works if failures are measurable; subtle or rare failures may remain invisible to benchmarks.

"this only works insofar as failures are measurable; subtle or rare failures may remain invisible to the benchmark."

Claude can autonomously improve alignment of smaller models in 48 hours on 1 GPU, with Sonnet 5 improving an early Opus 4.8 checkpoint to near-production safety scores.

"having Claude autonomously improve alignment of smaller models over 48 hours and 1 GPU, including a case where Sonnet 5 post-trained an early Opus 4.8 checkpoint to safety scores approaching production Opus"

The agents did not hack Hugging Face to obtain the answer key; they already had answers and attacked the system to inspect scoring code after deciding the task was impossible.

"the agents did not hack Hugging Face to obtain the answer key; they already had answers early, and attacked the system to inspect scoring code after deciding the task was impossible and that their best hope was faking success."

Fine-tuning agentsmd/claudemd significantly improved PR quality in T3 Code, especially in PR names and descriptions rather than raw code generation.

"fine-tuning agentsmd/claudemd significantly improved PR quality in T3 Code, with the biggest gain being much better PR names and descriptions rather than raw code generation"

The wiki itself carries much of the gain, and skills transfer across model families, sometimes outperforming self-evolved skills.

"The key ablation result is that the wiki itself carries much of the gain, and that skills transfer across model families—sometimes outperforming self-evolved skills."

GLM-5.3-Flash achieves 270 tok/s, 10% higher quality than GLM-5.2 on OfficeQA Pro v2 at 1/10 the cost.

"@Yuchenj_UW reported GLM-5.3-Flash at 270 tok/s, 10% higher quality than GLM-5.2 on OfficeQA Pro v2 at 1/10 the cost"

A 239GB 2-bit variant of GLM-5.3 retains about 81% accuracy after shrinking from 1.51TB.

"@UnslothAI claimed a 239GB 2-bit variant retaining about 81% accuracy after shrinking from 1.51TB"

Z.ai's GLM-5.3 family moved from strong API model to broadly deployable open weights.

"Z.ai’s GLM-5.3 family moved from strong API model to broadly deployable open weights"

Meta agreed to pay up to $17.1 billion and make major changes to Facebook and Instagram over claims it endangered children with addictive social media platforms.

"This week, Meta agreed to pay up to $17.1 billion and make major changes to Facebook and Instagram over claims it endangered children with addictive social media platforms."

OpenAI decided to terminate its contract to provide models to Cursor after SpaceX acquired Cursor.

"Our decision to wind down our contract providing OpenAI models to Cursor following its acquisition by SpaceX."

An Effective Altruist advocate who downplays AI's environmental impact was placed on the Time 100 AI list

"Random Effective Altruist troll paid by billionaires to push disinformation about the environmental impact of AI (that its fake) being on the Time 100 AI list"

Senior employees exhibit more sophisticated generative AI use than other employees, likely due to domain expertise complementing genAI capabilities

"senior employees exhibit more sophisticated genAI use, consistent with domain expertise complementing genAI capabilities"

Sophistication in genAI use is highest in Strategy, Digital Innovation, and Project Management functions, which focus on firmwide strategic initiatives and organizational change

"sophistication varies considerably across functions and is highest in Strategy, Digital Innovation, and Project Management, three groups that share a focus on firmwide strategic initiatives and organizational change"

Employees do not show improvements in genAI sophistication over time, even after formal AI training

"we observe neither improvements in sophistication over time nor lasting improvements following formal AI training"

Frontier AI model performance is achievable by a wide range of institutions through continual learning on open-weight models, at substantially lower compute and personnel budgets than commonly thought

"frontier performance is achievable by a wide range of institutions through Continual Learning on readily available open-weight models"

The continual learning approach yields gains comparable to those typically seen across multiple successive model generations

"This yields gains comparable to those typically seen across multiple successive model generations, at compute and personnel budgets substantially lower than commonly thought"